The tenant oplog applier doesn't grab the RSTL via AutoGetOplog before reserving oplog slots here, the way it does elsewhere. This needs to be added.

Currently, it's possible for the primary (on which the tenant oplog applier exists) to stepdown, but have the applier continue to run. This may cause a clash with the secondary oplog fetcher, and cause the stable timestamp to move ahead of the all durable timestamp, triggering an fassert.